> If you're using dumb boxes to deploy something like a k8s cluster why do you care about drive failure ? You rent out an extra machine to account for the downtime (and peaks) and you're still well below AWS infra cost.
k8s doesn't mean you don't care about data integrity or performance, and if you don't build and debug more automation yourself dealing with failures creates more manual labor recovering from them. Again, you can definitely do that yourself but it adds time commitments which blow the cost savings away unless you are running at least hundreds servers to have the savings exceed the staff cost.
> Again cattle not pets.
That's an orthogonal concern — neither environment will prevent you from treating servers as pets, and not having things like auto-scaling or higher-level service abstractions definitely encourages pet servers because the overhead of managing a full server and improving utilization adds up.
> But that's what I am saying - going "all in on Hetzner" doesn't really mean much if you're using dumb blocks available elsewhere (plain container/VM hosting, network object storage).
Actually, it does for the reasons I mentioned before. If you care about reliability, performance, security, etc. you're not just picking any random hoster around the world — you have to spend time selecting vendors, signing contracts, getting set up with their environment and tools, etc. Again, that's totally possible and there's a break-even point where those costs are canceled out by lower service costs but it's higher than people typically assume.
> (plain container/VM hosting, network object storage)
This is true but also means you need to factor in the cost of greater maintenance commitments and lower productivity.